Founder https://xitter.cf/t.co/xk5uayMLlP β the AI cloud and context layer for your team + https://xitter.cf/t.co/2niCosxp2o β context infrastructure for agents
π
Joined October 2014
- Tweets2.5K
- Following1.7K
- Followers3.7K
- Likes7.8K
New in Fabric:
Saw lots of photographers adopting Fabric, so we've just added support for all major camera (RAW) formats
Share, publish, review, straight from the camera
While everyone races to build the same thing... Fabric is thinking differently
The first AI... that plants trees π³
From today, every subscriber helps Fabric plant mangrove trees in Brazil and Kenya.
Mangrove trees are amazing because they protect shorelines, store carbon, and shelter marine life.
Good tools leave things better than they found them, and that's what we aspire to be at Fabric.
a well-executed idea and an ok one are absolutely galaxies apart
if it's not in the company's top 3 priorities, their execution will only ever be ok
can't recommend enough working on a product that you use every day, incredibly fun
(and yes, I've saved almost 15k things to this one Fabric folder)
evening project:
made the landing page for Fabric's handwriting scanner into an interactive mini-game
page is still WIP, but you can try it here: fabric.so/features/handwritiβ¦
to state the obvious here, calling current models "AGI" is disingenuous
they're extremely good at in-distribution tasks (e.g. coding), and terrible at out-of-band tasks, even if laughably simple like counting letters in a word, and even when pushed to reason it through
general intelligence != a high score on many trained formats
the latest models are superhuman within distribution and beyond brittle outside of it
taking the opposite side of my own point: this doesn't matter for tons of applications
a truly great coding model can build intellectual prosthetics on-demand to solve for its own limitations
to state the obvious here, calling current models "AGI" is disingenuous
they're extremely good at in-distribution tasks (e.g. coding), and terrible at out-of-band tasks, even if laughably simple like counting letters in a word, and even when pushed to reason it through
general intelligence != a high score on many trained formats
the latest models are superhuman within distribution and beyond brittle outside of it
making these spin in opposite directions is probably one of the best neuroflex exercises
very satisfying when it snaps into place
working on a little animation for the Fabric landing page
one fun detail: all the characters here are actually glyphs in a custom font
Full research paper, results and methodology here:
exabase.io/research/exabase-β¦
